Hi Dennis,

Thanks a lot for your feedback!

So I ended up using a direct solver, which works fine for my problem (2 phases, 
around 500k cells).

Best regards,
Nikolai


From: Dumux <[email protected]> On Behalf Of Dennis Glaser
Sent: Friday, August 31, 2018 20:27
To: [email protected]
Subject: Re: [DuMuX] Linear solver crash


Hi Nikolai,

I think the solver chosen in the exercise is not optimal for the matrix 
structures you have in the coupled fracture problem. We are still trying to 
find better solution strategies, however, in dumux/linear/seqsolverbackend.hh 
you can find two solvers that are designed for matrix structures like the one 
you have:

BlockDiagILU0BiCGSTABSolver

BlockDiagAMGBiCGSTABSolver

You can try these although I wouldn't put too much hope in it solving your 
problem. You can always use a direct solver if your system is not too big, i.e. 
the UMFPackBackend or the SuperLUBackend. These should always work but might 
need a lot of memory and much more cpu time. You need to have umfpack and/or 
superlu installed though.

I hope this helps!

Best wishes,
Dennis



On 31.08.2018 16:13, Nikolai Andrianov wrote:
Dear DuMuX experts,

While working on a modification of the 
dumux-course/exercises/exercise-fractures test case I observed that the linear 
solver crashes if the mesh is too fine. This results in cutting the Newton's 
time step, which in my case however does not help either..

This behavior is illustrated on the 
https://git.iws.uni-stuttgart.de/andrian/rate-sensitivity project on github: 
depending on the mesh density specified in the .geo files grids/fracs_bad.geo 
or grids/fracs_ok.geo, the linear solver either crashes, or the simulation 
proceeds as expected.

Your feedback is greatly appreciated.

Many thanks,
Nikolai





_______________________________________________

Dumux mailing list

[email protected]<mailto:[email protected]>

https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

_______________________________________________
Dumux mailing list
[email protected]
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

Reply via email to